Unit Leader @RIKEN_CBS | @RIKEN_PRI Investigating the computational principles of learning. Modeling synaptic plasticity and emergent neural dynamics.
Japan
Joined March 2024
- Tweets66
- Following149
- Followers364
- Likes61
How does a place cell know when to rewrite itself?
Check out our new preprint on spike-history gating of BTSP and closed-loop rewriting of CA1 representations. Congrats Anno! @ac_kurth
biorxiv.org/content/10.64898…
Toshitake Asabuki retweeted
Why did RNNs fail to learn long-term dependencies? What if we added one more modification, maybe now?
It turns out we can give a pretty broad, analytical answer! See the attached paper for a rigorous treatment using centre manifolds, low-rank RNNs, and dynamical systems theory!
Why do neural networks undergo abrupt learning after training on memory tasks? New research traces this phenomenon to a “ghosting” mechanism that locks the system into a no-learning zone and proposes two practical fixes to address these pitfalls.
📝 go.aps.org/4fXWEeF
ALT Panel B displays a series of four three dimensional graphs plotting the progression of speed minima in the neural activities of full-rank recurrent neural networks (RNNs) performing a DCD task. PC1 is on the x-axis, PC3 is on the y-axis, and PC2 is on the z-axis. The four graphs correlate to Epoch 11, Epoch 201, Epoch 1001, and Epoch 3001, from left to right. The speed minimum is denoted with diamonds. Trial type is denoted with F for fixed-point, S for slow-point, and G for ghost. Cue 1 is denoted with a blue line, and Cue 2 with an orange line. Panel D displays similar graphs but for an RNN performing an DMTS task. Here, match is colored with blue and orange lines, and no-match is colored with green and red lines. On the right, a color scale denotes the normalized speed of the fixed, slow, or ghost points increasing from dark blue to dark red.
Thrilled to share our new preprint 📜
Predictive E/I plasticity can drive intrinsic representational drift while preserving stable population coding.
w/ @ClopathLab
biorxiv.org/content/10.64898…
Toshitake Asabuki retweeted
More on Brain Waves! How does the brain coordinate information flow across widespread networks? New research reveals that traveling waves aren't just localized phenomena—they are a distributed organizational principle. By combining wide-field imaging and high-density recordings, Ye et al. show how the literal geometry of axonal connections dictates the direction and timing of brain-wide rotating waves during behavior.
science.org/doi/10.1126/scie…
#Neuroscience #SciencePaper #ComputationalNeuroscience #NeuralDynamics #BrainResearch
#BrainWaves #Brain #Biology @ScienceMagazine
How can RNNs combine independently learned computations without being trained on their combinations?
Excited to share our new preprint📜
biorxiv.org/content/10.64898…
Toshitake Asabuki retweeted
Excited to share our new findings on how internal states dynamically modulate the population dynamics!
This was only possible due to the amazing mentorship of @SurLabMIT and @timbuschman!
Check it out here!
biorxiv.org/content/10.64898…
How do recurrent networks decide what to spontaneously replay? Check out our new preprint!
w/@ClopathLab
biorxiv.org/content/10.64898…
Toshitake Asabuki retweeted
How does the brain build a memory?
A common assumption is that the neurons activated during an experience collectively form the memory engram.
In our new Nature Neuroscience paper (finally out!), we show that this is not the case.
nature.com/articles/s41593-0…
Toshitake Asabuki retweeted
Important clarification that they used a classifier to connect simulated neural activity patterns onto 3D movements. You don't need a connectome (or even brain-like activity) for that. Any semi-structured sequential pattern will do
There's a fruit fly walking around right now that was never born.
@eonsys just released a video where they took a real fly's connectome — the wiring diagram of its brain — and simulated it. Dropped it into a virtual body. It started walking. Grooming. Feeding. Doing what flies do.
Nobody taught it to walk. No training data, no gradient descent toward fly-like behavior. This is the opposite of how AI works. They rebuilt the mind from the inside, neuron by neuron, and behavior just... emerged. It's the first time a biological organism has been recreated not by modeling what it does, but by modeling what it is.
A human brain is 6 OOM more neurons. That's a scaling problem, something we've gotten very good at solving. So what happens when we have a working copy of the human mind?
Toshitake Asabuki retweeted
Online now: Linking neural manifolds to circuit structure in recurrent networks dlvr.it/TRL6b7
AI forgetting is a “traffic jam” inside the network. 🚦
In #ContinualLearning for RNNs, shared weight update directions overwrite learned dynamics.
We show that simple feedback switching guides plasticity into task-specific manifolds.
With @Lewieliu987, @ac_kurth, @OsakoYuma
How can RNNs learn continuously without forgetting? 🧠
Our new preprint shows how a predictive learning rule organizes recurrent dynamics into orthogonal manifolds, reducing task interference.
Congrats Zihan @Lewieliu987!
biorxiv.org/content/10.64898…
You can find the full preprint and more details on our approach here:
biorxiv.org/content/10.64898…
How can RNNs learn continuously without forgetting? 🧠
Our new preprint shows how a predictive learning rule organizes recurrent dynamics into orthogonal manifolds, reducing task interference.
Congrats Zihan @Lewieliu987!
biorxiv.org/content/10.64898…
Finding 5: Even in a 500-unit recurrent network generating 218k-pixel movie frames, geometric isolation persists, enabling rapid re-learning through reuse of previously learned dynamics.
Finding 2: The degree of manifold orthogonality predicts re-learning speed, consistent with experimental observations (Kim et al., Nature 2025).
Toshitake Asabuki retweeted
Our paper has been published! Congratulations to Kiyooka-san, Oomoto-san, and the team!
📃Single-cell resolution functional networks during unconsciousness are segregated into spatially intermixed modules
doi.org/10.1016/j.celrep.202…
新作論文：意識・無意識脳での神経ネットワーク構造の可視化に成功-睡眠中に感覚応答を知覚できない脳の謎にヒント-ハブ細胞の役割とは？大泉ラボ@oizumimとのコラボ。同じデータでマクロからミクロまでネットワーク指標を解析するパイプラインを構築
doi.org/10.1016/j.celrep.202…
riken.jp/press/2026/20260207…